h t t p s : / / g i t h u b . c o m / s p a c e d r i v e a p p / s p a c e b o t

Spacebot: AI Agent for Teams and Communities

Type: web-page

Author: spacedriveapp Source: https://github.com/spacedriveapp/spacebot

Repository: spacedriveapp/spacebot Description: An AI agent for teams, communities, and multi-user environments Stars: 1.2k | Forks: 154 License: FSL-1.1-ALv2 Language: Rust Website: spacebot.sh

Overview

Spacebot is a concurrent AI agent framework designed for multi-user environments. Rather than processing everything sequentially in a single session, it splits functionality across specialized processes that work in parallel—channels handle conversation, branches manage thinking, workers execute tasks, and a cortex supervises the entire system.

Core Architecture

The system uses five process types:

  • Channels: User-facing LLM processes that never block, always responsive to messages
  • Branches: Independent thinking forks that inherit channel context for analysis
  • Workers: Specialized task executors (shell, file, browser, coding) spawned by branches
  • Compactor: Programmatic monitor that prevents context overflow without interrupting conversation
  • Cortex: System-wide supervisor managing memory, process health, and knowledge synthesis

Key Capabilities

Task Execution: - Shell commands with configurable timeouts - File operations (read, write, list with auto-creation) - Program execution with environment variables - Headless browser automation with accessibility trees - OpenCode integration for deep coding sessions - Brave web search integration

Messaging: Message coalescing detects rapid-fire bursts and batches them into single LLM turns. Native adapters support Discord, Slack, Telegram, Twitch, and Webchat with rich formatting, threading, reactions, and per-channel permissions.

Memory System: Eight typed memory categories (Fact, Preference, Decision, Identity, Event, Observation, Goal, Todo) with graph edges (RelatedTo, Updates, Contradicts, CausedBy, PartOf). Hybrid recall merges vector similarity and full-text search via Reciprocal Rank Fusion. Memory bulletin provides periodic briefings injected into conversations.

Model Routing: Four-level system selecting appropriate models per call—process-type defaults, task-type overrides, prompt complexity scoring, and fallback chains. Supports Anthropic, OpenAI, OpenRouter, Z.ai, Groq, Together, Fireworks, DeepSeek, xAI, Mistral, NVIDIA, MiniMax, Moonshot AI, Ollama, and custom OpenAI/Anthropic-compatible endpoints.

Extensibility: - MCP (Model Context Protocol) integration for arbitrary tool access - skills.sh registry for installing community skills - Hot-reloadable configuration for live updates

Scheduling: Cron jobs with natural language creation, clock-aligned intervals, active hours, per-job timeouts, and circuit breaker auto-disabling after failures.

Deployment Options

  • spacebot.sh: One-click hosted deployment
  • Self-hosted: Single Rust binary, no dependencies
  • Docker: Container with volume persistence

Getting Started

Requirements: Rust 1.85+ and an LLM API key

git clone https://github.com/spacedriveapp/spacebot
cd spacebot
cargo build --release

Minimal configuration connects Discord, Slack, or other platforms with routing to any supported LLM provider.

Design Philosophy

"Thinks, executes, and responds — concurrently, not sequentially. Never blocks. Never forgets." The architecture prevents the bottlenecks found in traditional single-threaded agent frameworks where conversation freezes during context compaction or task execution.